They even discuss it in detail:
"What makes NeighborhoodScout® Crime Data uniquely accurate?
Most city neighborhood crime data are incomplete and inaccurate because crimes are reported by individual law enforcement agencies, rather than by city or town, and many cities – even small ones – have more than one agency responsible for law enforcement (municipal, university, county, transit, etc.). Even FBI data are reported by agency not by city or town, providing an incomplete assessment of city-wide crime counts. It is an agency-centric rather than locality-centric reporting method. If you use FBI data, you only get city-wide general counts, and only from one agency in the city, so it is generally incomplete for the city overall, as well as not specific to a neighborhood or address."
And further:
Once we have these complete set of reported crime data, along with millions of geocoded reported crime incidents using a GIS, we begin our crime data development process.
The results are fine resolution, highly accurate crime data that are comparable nationally.
Our approach provides you the ability to look at small areas effectively.
In some cases a city agency is in charge of law enforcement, while in other areas it’s a county. In many cases it is more than one agency for a geographic area. Since the geography varies, it’s difficult to compare the scores among jurisdictions, or to get a true and complete picture of crime risk. This is why we use a relational database to assess the true count of reported crimes in a locality.
Although most agencies report, not all do. This creates holes in the data. Our method allows us to accurately fill in the holes based on the crime experience of many like locales, and provide accurate crime data for anywhere in the U.S.